Philadelphia, PA—A pragmatic review of literature has identified best practices for network meta-analyses (NMAs) of disease-modifying drugs for the treatment of relapsing-remitting multiple sclerosis (RRMS).
“Based on our critical appraisal,” according to Amy Phillips, PharmD, EMD Serono Inc, Rockland, MA, and colleagues, “best practices for NMAs in RRMS include considerations for end points, analytical approaches, potential subgroup and meta-regression analyses, and tests for consistency and sensitivity.”
The study was presented at a poster session at the International Society for Pharmacoeconomics and Outcomes Research 2015 annual meeting. NMAs provide a valuable means for assessing indirect evidence, but analyses vary in their inclusion criteria, methodology, and types of statistical syntheses. Therefore, the researchers noted, a critical assessment of methodological approaches is needed when evaluating results.
“Most NMAs fail to provide a definitive basis for clinical guidance,” Dr Phillips and colleagues noted. “This is in part because studies are conducted over limited periods, whereas the effects of new treatments are generally unclear beyond 2 years in a therapeutic area that may extend up to 40 years in duration.”
According to researchers, a consensus array of inclusion and exclusion criteria, end points, and statistical models would help in synthesizing evidence to guide clinical decisions in RRMS.
Using key search terms, Dr Phillips and colleagues scanned databases, including PubMed/Cochrane Library, key scientific congresses, and relevant health technology assessment organizations. They then assessed and extracted the central features of each identified NMA for analysis.
The researchers identified 25 NMAs as relevant. According to the investigators, randomized controlled trial and RRMS were the most frequently cited inclusion criteria in the literature reviews. “In 5 of the 25 studies, the patient population would be considered if greater than 80% of patients had RRMS,” she reported. “In 1 study, the patient population would be considered if greater than 50% had RRMS.”
Annualized relapse rate was the most frequently reported end point, closely followed by percentage of patients with disease progression. “End points that are most relevant to patients with RRMS should be explored in any efficacy-focused analysis,” the researchers commented. “These end points should be investigated at a 2-year or longer follow-up duration, if possible, given the slow progression of RRMS.”
Most NMAs investigated self-injectable therapies as comparator strategies, including interferon beta-1a (Rebif, Avonex), interferon beta-1b (Betaseron), and glatiramer (Copaxone).
A frequentist random-effects approach was the most common analysis framework (used in 40% of studies), although Bayesian methodology has become more common in recent NMAs (used in 28% of studies), researchers observed. “The selection of the base-case analysis approach should be driven by the desired model outputs,” noted Dr Phillips and colleagues. “But any analysis should aim to compare findings with those produced by a frequentist framework as a sensitivity analysis to evaluate the robustness of findings.”
Although similar end points are regularly compared across studies, according to the researchers, the criteria for classifying these end points can vary. In addition, differences in patient demographics are likely to make an impact on the treatment effects observed in any given trial.
“Therefore, models that use a random effects approach are likely to have greater face validity than those using
a fixed-effects model, as the random-effects model better reflects the likely heterogeneity of treatment effects across studies,” Dr Phillips and colleagues noted.
Baseline disease characteristics and publication year also had a significant impact on results, probably driven by changes in the diagnostic criteria over time. Thus, meta-regression analyses may be useful within the RRMS space, as study-level differences could lead to significant differences in results observed across trials.
Finally, the researchers indicated that although consistency approaches were not frequently reported, they should be included to support the validity of results. “The majority of NMAs did not report sensitivity analyses,” they noted. “However, these should be performed across model methodologies to ensure the robustness of results.”
Last modified: August 5, 2015